On Thu, Oct 29, 2020 at 6:56 AM Jan Kiszka <[email protected]> wrote:
> On 29.10.20 11:49, Renato Mancuso wrote: > > Hi Jan, Andrea, > > > > Thanks for looping me in on this topic! > > > > Just a not that I in my version I don't just recreate the root page, but > > the entire page table tree to be NC for the SMMU. That immediately gave > > me a stable system on the ZCU102. I have tested that even with a > > dynamically recolored root cell. I haven't tested with colored > > DMA-capable inmates, but I'll do that soon as I have just added support > > to boot colored Linux inmates. I keep SMMU and CPU tables in sync by > > duplicating any change in map/unmap operations at cell creation time. > > That recreation is what we do for VT-d vs. VT-x as well because tables > are incompatible (at least in the VT-d version we support). On AMD, that > is not needed, and I thought that the ARM tables were designed to enable > the same. How does Linux handle this? > On ARM the format of SMMU page tables is supposed to be compatible. Re-using the CPU page tables is also what Linux does for SMMUv2. However as far as I can tell there is a lack of testing for the SMMUv2 support of Linux on the ZCU102. It might as well be an errata of the specific instance of SMMU on my cut of ZCU102 systems. I know that Andrea has been performing some more debugging to test this hypothesis. -- Renato -- You received this message because you are subscribed to the Google Groups "Jailhouse" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/jailhouse-dev/CAL30Xq_jo2vLwtVdAC_nFHvqdMjs_wmJT5quqz3Xb1Op8217FA%40mail.gmail.com.
